基于改进ART2神经网络的发动机故障诊断方法

Engine Fault Diagnosis Method Based on Modified ART2 Neural Network

基于改进ART2神经网络的发动机故障诊断方法,用警戒和调整因子的双因子法控制网络识别过程中对已知故障再学习,使网络不断学习和优化。以某发动机的相关状态模式训练ART2网络,利用db4小波包对各模式的振动信号进行分解,再利用小波系数计算出各频带的能量构成向量,经归一化后为该模式下的特征向量。其网络只对相似度超过调整因子的识别样本进行学习,有助于提高网络发动机状态模式的识别精度。

Engine fault diagnosis method based on modified ATR2 neural network, by using caution and adjustment factor-the method of double factors, the relearning process of known fault pattern was controlled, and the network was in the condition of continuous learning and optimization. Some relative condition patterns of a certain engine were used to train ART2 net. All vibration signals in every mode were dissolved with db4 wavelet packet. And then taking use of wavelet parameters, the energy vector of each frequency was calculated. After normalization, the characteristic vector in this pattern was acquainted. This net only studies such sample that its similarity exceeds adjustment factor and this would enhance the precision of recognition.